liveBook · Manning

Reliability

As explained in The Coder Cafe

Reliability is the ability of a system to continue performing its intended functions correctly over time and under expected failure conditions.

Reliability is the user-facing promise that a system will do what people expect, consistently, over a specified period. “Merged and deployed” is not the same as “available and working.” Reliability matters even when a failure does not cause a crash: lost data, an unavailable service, or an unsafe result can each break that promise. section 7.3.4 section 7.0

What counts as reliable?

A cloud datastore might promise availability 99.99% of the time: availability states what users expect, while the percentage makes the time-based promise measurable. A watch tested over several days for precise timekeeping makes the same distinction. The stakes vary: an air-traffic-management failure may put lives at risk, while losing years of cloud-stored family photos can destroy irreplaceable memories. section 7.3.4 section 7.1.1

Reliability is the result, not a synonym for every quality that supports it. Resilience concerns recovery; fault tolerance maintains continuity through redundancy; robustness withstands stress. Performance and scalability can also help deliver a consistently working service. The useful question is which qualities the user’s promise requires. section 7.8

Make failure behavior a design decision:

  • Degrade gracefully when full service is impossible. Load shedding or adaptive LIFO can preserve service for some users instead of failing everyone. section 7.8
  • Choose whether an operation fails open or fails closed. The choice balances availability against control and security; it depends on the cost of an unsafe operation versus an unavailable service. section 7.8
  • Classify dependencies. A hard dependency must be at least as reliable as the service relying on it. A soft dependency needs an actively maintained, continuously tested fallback so it stays soft. section 7.8

Reliability also depends on what happens after an incident. A change preceding a failure is not necessarily its cause. Look for evidence that could disprove the leading hypothesis, and segment metrics by factors such as platform, region, or cohort to expose hidden influences. Dashboards with useful breakdowns can shorten the search; a postmortem can identify breakdowns to add next time. section 7.8 section 7.7.5

Testing can express a reliability promise as a system-level property. For example, retrying a failed request must not create a duplicate financial transaction. Property-based testing can check such invariants with randomized inputs across an API or interacting applications; it complements rather than replaces traditional tests. section 5.2.4

How The Coder Cafe covers it

How deep it goes
Defines a time-bound user expectation and outlines failure-design and debugging practices without prescribing one implementation.
Examples
Illustrates the stakes with air traffic and lost cloud photos, then makes expectations concrete with watch precision and datastore availability.
What it assumes
Assumes familiarity with software systems while defining reliability and distinguishing it from related system qualities.
What you can do after
Practice/perform reliability design and incident diagnosis through dependency classification, graceful degradation, and segmented metrics.
How it connects
Connects resilience, fault tolerance, robustness, graceful degradation, dependency management, and root-cause analysis.

Also explained in

Compare how 21 books explain reliability →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of The Coder Cafe. The section links point to the parts of the book it draws on.